The accurate measurement of semiconductor substrate temperatures during processing is highly desirable for semiconductor substrate processing. In particular, most processes are temperature sensitive and therefore, accurate temperature measurement is a pre-requisite to the control of optimal conditions for etch and/or deposition chemistry. Moreover, a spatial variation of temperature across a semiconductor substrate can lead to non-uniform processing when either etching or depositing material.
There are three geometric modes or configurations of band-edge thermometry (BET): (1) transmission spectroscopy (TS; see FIG. 1C), (2) specular reflection spectroscopy (SRS; see FIG. 1D), and (3) diffuse reflectance spectroscopy (DRS; see FIGS. 1A and 1B). The geometry of each mode is presented in FIGS. 1A-1D.
In the DRS mode, the light source and detector are on the same side of the substrate with the detector placed in a non-specular position (see Johnson et al., U.S. Pat. Nos. 5,568,978 and 5,388,909 (hereinafter “the '978 patent” and “the '909 patent,” respectively)). A non-specular detector only sees the light that is transmitted through the wafer and that is diffusely back scattered into the solid angle of the detector. In the DRS method, the double-pass transmission of light through the substrate is measured as a function of wavelength or, equivalently, photon energy. As the wavelength increases, the photon energy decreases, and the onset of substrate transparency occurs as the photon energy becomes less than the band-gap energy.
In the SRS mode, the light source and detector are also on the same side of the substrate. The detector is placed in a specular position where it detects light that is specularly reflected from both surfaces of the wafer (see Cabib & Adel, U.S. Pat. No. 5,322,361 (hereinafter “the '361 patent)). The light that is reflected into the detector without traveling through the wafer contains no temperature information and consequently adds only a relatively constant background signal. The light component that is reflected from the opposite internal surface of the substrate travels back through the wafer and onto the detector. That reflected component, which passes twice though the wafer, contains the useful temperature information.
In the TS mode, the onset of substrate transparency (or, equivalently, the band-gap energy) is determined by the transmission of light through the substrate as described in Kirillov & Powell (U.S. Pat. No. 5,118,200 (hereinafter “the '200 patent”)). In this geometry, the light source and the detection system are on opposite sides of the wafer. One difficulty with this approach is that it requires optical access to the chamber at opposite sides of the substrate. However, in comparison to the SRS mode, the TS mode results in an increase in the light intensity received by the optical detector.
No matter what mode is used, a temperature signature must be extracted from the spectra. In general, three algorithms have commonly been used to extract substrate temperature from band-edge spectra: (1) the spectral position of the maximum of the first derivative or, equivalently, the inflection point, (2) a direct comparison of the spectrum to a predetermined spectral database, and (3) the position of the spectrum knee (i.e., the location of the maximum of the second derivative). The first method has been discussed in the '200 patent. That method determined the substrate temperature as a function of the position of the inflection point of the spectrum in a previous calibration run where the temperature of each spectrum is known. The advantages of that method are that it is simple, fast and independent of the absolute intensity of measurement. The disadvantage is that it is very sensitive to interference effects that may occur at either surface of the processed silicon (Si) wafer.
In the second approach, the '361 patent compares a given spectrum to a temperature-dependent database composed of spectra taken at known temperatures. One advantage is that it is reported to work well for Si wafers. A disadvantage is that it is sensitive to interference effects and requires an absolute reflectivity measurement. Accordingly, each wafer may require a separate normalization spectrum.
Lastly, the '978 and '909 patents disclose a DRS mode BET, using the position of the spectrum knee as a signature. Its advantage is that it is the closest distinct point to the onset of transparency of a substrate, and is therefore less sensitive to interference effects. A shortcoming of this approach is that it requires sophisticated fitting algorithms that may be too slow for some current applications.
In general, a BET system includes three main units, i.e., a light source, a dispersion device and a photo-detector. Currently, there are several commercially available systems; however, none of these systems is fully capable of the following criteria:
1) Non-contact thermometry from the bare backside of Si wafers during front side processing.
2) Use of optical methods and quartz rods to couple light in and out of the process chamber.
3) Two-dimensional snapshot of wafer temperature.
4) Simultaneous samples of several points (approximately 10) on large Si wafers with a response time of 100 msec or less.
5) Temperature range of 20 to 300° C.
6) Accuracy of temperature measurement to within 2 to 5° C.